Featuretheorems – after all, the HOL Light and Isabelle Librariescontain Tom Hales’ proof of the Kepler Conjecture and theCoq library contains the proof of Feit-Thomson’s Odd-OrderTheorem.Note that once a set of views has been found by the viewfinder,we can generate the induced theorems in all accessibletheories and iterate the view finding process, which might findmore views with the induced theorems.Figure 5. Searching for Induced Knowledge Itemsgraph – contains enough information to generate an explanationof the reason + is associative on Z. In essence, thisexperimental search engine searches the space of (induced)mathematical knowledge rather than just the space of mathematicaldocuments.Making Bourbaki AccessibleIdeas like the ones above can solve one of the problems withthe Bourbaki book series, which is written in such a very conciseand modular manner that it can only be understood if onehas all the previous parts in memory. We have extracted thetheory graph underlying the first 30 pages of Algebra I [2].It contains 51 theories, 94 inclusions and 10 views. The theoriescontain 82 Symbols, 38 axioms, 30 theorems and 17definitions [13]. For knowledge items higher up in the graph,there is no (single) place in the book which states all theiraxioms or properties. With the MMT API, we can generateflattened descriptions for reference and with FlatSearch wecan search for their properties. We conjecture that the simpleexplanation feature from Figure 5 can be extended into a“course generator” that generates a self-contained – modulothe reader’s prerequisite knowledge – document that explainsall aspects necessary for understanding the search hits. Withtheory graph technology, Bourbaki’s Elements can be read byneed, as a foundational treatise should be, instead of beingrestricted to beginning-to-end reading.Uncovering TheoremsWe can now come back to the discussion on the OBB from theintroduction. One way to overcome the problem of missinginterpretations is to systematically search for them – after all,machine computation is comparatively cheap and the numberof potential theory morphisms is bounded by the squareof the number of theories times the number of symbols intheories. In particular, the MathWebSearch formula searchengine [10] can efficiently search for substitutions (which areessentially the same as the symbol-to-expression mappingsof theory morphisms). We have explored this idea before thetheory graph technology was fully developed, and the Theo-Scrutor system [17] found a considerable number of simpleviews in the Mizar library. While these were relatively syntacticand obvious – after all, for a view finder to work, theproofs for the proof obligations have to be part of the libraryalready – they had not previously been noticed because of theOBB, even though the Mizar project has a “library committee”tasked with finding such linkages. We expect that oncethe OAF is sufficiently stable, a renewed experiment will yieldmany more, and more interesting views and possibly novelRefining Theory GraphsFinally, even partial views – which should be much more numerousthan total ones in a theory graph – can be utilised. Saywe have the situation below with two theories S and T, a partialtheory morphism S −→ σT with domain D and codomainC, and its partial inverse δ.SDσδTCThen we can pass to the following more modular theorygraph, where S ′ := S \D and T ′ := T\C. In this case wethink of the equivalent theories D and C as the intersection oftheories S and T along σ and δ. Note that any views out of Sand T now have to be studied, if they can be pulled back to Cand D.S ′ T ′δσσDCδWe have observed that many of the lesser known algebraicstructures in Bourbaki naturally arise as theory intersectionsbetween better known structures. We hope to explain the remainingones via other category-theory-inspired theory graphtransformation operations.Note that operations like theory intersections apply theorygraph technology to the problem of theory graph maintenance,which is itself a problem greatly hampered by the OBBwithout MKM techniques.5 ConclusionIn this paper we have explored opportunities to lift the onebrainbarrier in mathematics, which limits the application ofmathematical knowledge both inside the mathematical domainas well as in other disciplines. We propose that the wayforward is to employ computer systems that can systematicallyexplore immense knowledge spaces, if these are representedin sufficiently content-oriented formats. Together withthe creation, curation and application of digital mathematicallibraries (DMLs), this is one of the central concerns of thenew field of Mathematical Knowledge Management (MKM).We have presented the theory graphs approach as a representationparadigm for mathematical knowledge that allowsus to make its modular and highly networked structure explicitand therefore machine-actionable. We have seen thattheory graphs following the “little theories” approach containthe information structures necessary to explain – and thus ultimatelysupport by computer – many mathematical practices26 EMS Newsletter June 2014
Featureand culture skills. This has the potential to significantly extendthe “mathematical literacy” of mathematical knowledgemanagement systems and consequently make them more suitableas tools that complement human skills.We have explored three exemplary mathematical applicationsof theory graph technologies and one MKM-internal togive an intuition of what services we can expect if we embarkon the enterprise of representing large bodies of mathematicalknowledge and its network structures in machine-actionableformats. The availability of such DMLs is currently the largestbottleneck for overcoming the OBB in mathematics. We arecurrently experimenting with establishing an open archive offormal mathematics (the OAF project [18]) by integrating theoremprover libraries. But formalisation often poses a highburden on the author and forces decisions about logical foundationsthat are irrelevant mathematically. Therefore, we arecurrently researching ways the theory graph methods presentedhere can be extended to representations of mathematicalknowledge in which the degree of formalisation is flexible.Flexiformal representations – see [9] for a discussion –are much closer to mathematical vernacular, which mixes informalparts (natural language) with formal parts (e.g. formulaeand functional markup for mathematical statements) andare therefore easier to obtain in practice. But mathematicalliteracy may be limited by the availability of formal/machineactionableparts; therefore, we are additionally investigatingmethods for automated semantics-extraction from mathematicaldocuments, which would greatly enhance the reach of themethods described in this paper.AcknowledgementsWork on the concepts presented here has been partially supportedby the Leibniz association under grant SAW-2012-FIZ_KA-2 and the German Research Foundation (DFG) undergrants KO 2428/9-1 and KO 2428/13-1.Bibliography[1] arxiv.org e-Print archive. URL: http://www.arxiv.org.[2] N. Bourbaki. Algebra I. Elements of Mathematics. SpringerVerlag, 1974.[3] W. M. Farmer. Mathematical Knowledge Management. InD. Schwartz and D. Te’eni, editors, Encyclopedia of KnowledgeManagement, pages 1082–1089. Idea Group Reference,2 edition, 2011.[4] W. M. Farmer, J. Guttman, and X. Thayer. Little theories.In D. Kapur, editor, Proceedings of the 11 th Conference onAutomated Deduction, number 607 in LNCS, pages 467–581,Saratoga Springs, NY, USA, 1992. Springer Verlag.[5] J. Harrison. Formal Proof – Theory and Practice. Noticesof the AMS, 55(11):1395–1406, 2008. URL: http://www.ams.org/notices/200811/tx081101395p.pdf.[6] J. Jeuring, J. A. Campbell, J. Carette, G. Dos Reis, P. Sojka,M. Wenzel, and V. Sorge, editors. Intelligent Computer Mathematics,number 7362 in LNAI. Springer Verlag, 2012.[7] A. Kohlhase and M. Kohlhase. Spreadsheet interaction withframes: Exploring a mathematical practice. In J. Carette,L. Dixon, C. Sacerdoti Coen, and S. M. Watt, editors, MKM/-Calculemus Proceedings, number 5625 in LNAI, pages 341–356. Springer Verlag, July 2009. URL: http://kwarc.info/kohlhase/papers/mkm09-framing.pdf.[8] M. Kohlhase. OMDoc – An open markup format for mathematicaldocuments [Version 1.2]. Number 4180 in LNAI.Springer Verlag, Aug. 2006. URL: http://omdoc.org/pubs/omdoc1.2.pdf.[9] M. Kohlhase. The flexiformalist manifesto. In A. Voronkov,V. Negru, T. Ida, T. Jebelean, D. Petcu, S. M. Watt, andD. Zaharie, editors, 14th International Workshop on Symbolicand Numeric Algorithms for Scientific Computing (SYNASC2012), pages 30–36, Timisoara, Romania, 2013. IEEE Press.URL: http://kwarc.info/kohlhase/papers/synasc13.pdf.[10] M. Kohlhase, B. A. Matican, and C. C. Prodescu. MathWeb-Search 0.5 – Scaling an Open Formula Search Engine. InJeuring et al. [6], pages 342–357. URL: http://kwarc.info/kohlhase/papers/aisc12-mws.pdf.[11] M. Kohlhase, H. Mihaljevic-Brandt, W. Sperber, andO. Teschke. Zentralblatt column: Mathematical formulasearch. EMS Newsletter, pages 56–57, Sept. 2013. URL: http://www.ems-ph.org/journals/newsletter/pdf/2013-09-89.pdf.[12] C. Lange, P. Ion, A. Dimou, C. Bratsas, J. Corneli,W. Sperber, M. Kohlhase, and I. Antoniou. Reimplementingthe mathematics subject classification (MSC) as alinked open dataset. In Jeuring et al. [6], pages 458–462.http://arxiv.org/abs/1204.5086 arXiv:1204.5086.[13] B. Laubner. Using theory graphs to map mathematics: A casestudy and a prototype. Master’s thesis, Jacobs University, Bremen,Aug. 2007. URL: https://svn.eecs.jacobs-university.de/svn/eecs/archive/msc-2007/blaubner.pdf.[14] Mizar. URL: http://www.mizar.org.[15] International Conference on Mathematic Knowledge Management.URL: http://www.mkm-ig.org/.[16] Mathematical reviews. URL: http://www.ams.org/mr-database.[17] I. Normann. Automated Theory Interpretation. PhD thesis,Jacobs University, Bremen, Germany, 2008. URL:https://svn.eecs.jacobs-university.de/svn/eecs/archive/phd-2008/normann.pdf.[18] The oaf project. URL: http://mathhub.info/OAF-project.[19] F. Rabe. The MMT API: A Generic MKM System. InJ. Carette, D. Aspinall, C. Lange, P. Sojka, and W. Windsteiger,editors, Intelligent Computer Mathematics, number7961 in Lecture Notes in Computer Science, pages 339–343.Springer, 2013. http://dx.doi.org/10.1007/978-3-642-39320-4doi:10.1007/978-3-642-39320-4.[20] F. Rabe and M. Kohlhase. Information & Computation,0(230):1–54, 2013. URL: http://kwarc.info/frabe/Research/mmt.pdf.[21] F. Wiedijk. Formal proof – Getting Started. Notices of theAMS, 55(11):1408–1414, 2008. URL: http://www.ams.org/notices/200811/tx081101408p.pdf.[22] Zentralblatt MATH. URL: http://www.zentralblatt-math.org/zbmath/.Michael Kohlhase [m.kohlhase@jacobsuniversity.de]is a full professor for computerscience at Jacobs University Bremenand an associate adjunct professorat Carnegie Mellon University. He studiedpure mathematics at the Universities ofTübingen and Bonn (1983–1989) and continuedwith computer science, in particularhigher-order unification and automated theorem proving(PhD 1994, Saarland University). His current researchinterests include knowledge representation for mathematics,inference-based techniques for natural language processingand computer-supported education.EMS Newsletter June 2014 27